To address the persistent challenge of nonlinear prediction associated with prestress loss in slope anchor cables within the field of anchoring engineering,this study proposes the implementation of a long short-term memory(LSTM) network augmented by a weighted average optimization algorithm(WAA).The WAA is utilized to optimize key parameters,including regularization parameters,initial learning rate,and the number of LSTM layer units,thereby developing a WAA-LSTM-based model for the prediction of long-term prestress loss.Empirical data on anchor cable prestress from the southwestern slope of an open-pit mine in Southwest China served as the dataset for training and testing the WAA-LSTM model. The model’s predictive performance was evaluated against measured values,traditional LSTM,backpropagation(BP) neural network,particle swarm optimization-LSTM(PSO-LSTM),and H-2K creep theory calculations.Furthermore,recursive prediction was employed to estimate the unknown prestress loss over the ensuing two months. The findings indicate that the WAA-LSTM model surpasses traditional LSTM,BP neural network,PSO-LSTM,and H-2K creep theory in terms of mean absolute error(MAE),mean absolute percentage error(MAPE),root mean square error(RMSE),and coefficient of determination(R²),with prediction deviations reduced by 19.7%~41.3% and the highest R² reaching 0.9847.Additionally,during the two-month prediction period for unknown values,the predicted prestress loss showed no significant deviation or abrupt changes from historical trends,demonstrating the model’s ability to effectively capture the periodic influence of dynamic factors on prestress loss. Some predicted curves exhibited slight fluctuations within stable intervals,aligning with the mechanical mechanism of time-dependent damage coupled with external disturbances.The results address the limitations of linear assumptions in traditional theoretical models and offer a high-precision predictive approach for long-term stability assessment in slope anchoring engineering.
目前,针对锚索预应力损失预测的研究主要基于蠕变耦合理论,建立通过岩体与锚索相互作用的力学耦合模型来推导预应力损失解析解的预测模型(吴小萍等,2023;杨文东等,2023)。通过这种方式,早期构建的理论预测模型主要是采用弹性体与广义Kelvin体并联的H-K模型(俞强山等,2019),但该模型主要适用于短期的预应力损失预测,在长期预测中精度不足。因此,后续研究主要通过改进H-K模型结构来提高长期预测精度(任青阳等,2020;冯忠居等,2021),如引入多个Kelvin体构建H-2K或H-3K模型,并在此基础上提出分阶段预测理论,这种方式极大地提高了锚索预应力损失预测的准确性(徐毅青等,2020;Wang et al.,2024)。同时,不同的锚固形式和岩体结构均会对理论蠕变预测模型的推导产生影响,如面对桩锚结构中的锚索预应力损失问题时,蠕变模型需同时考虑锚索—岩土耦合蠕变行为以及锚固装置和腰梁变形,使用整数阶导数方法描述耦合蠕变行为,推导蠕变方程和松弛方程(Gao et al.,2022)。然而,在土岩双结构边坡等复杂地质条件下,锚索预应力损失往往呈现“快速下降—回升—趋稳”的三阶段非线性特征。为描述这种变化趋势,研究人员主要通过回归分析与分段建模的方式来构建和改进耦合蠕变预测模型(Gao et al.,2021)。
现有研究中,虽然各种理论预测模型均能够反映材料的时变特性,但其应用仍存在明显的局限性:一方面,模型参数的确定依赖于特定工程的监测数据来反演,普适性较差;另一方面,这些模型对降雨和温度波动等非线性外部因素的适应性较弱,难以准确模拟实际工程环境下的预应力演化规律。因此,探究新的锚索预应力损失预测理论和方法显得很有必要。近年来,机器学习方法在矿山安全工程领域得到了成功应用,如巷道围岩松动圈预测、采场稳定性分析、地面沉降监测和露天矿爆破振动峰值速度预测等(谢饶青等,2022;方博扬等,2023;李振阳等,2024;王本浩等,2024)。其中,长短期记忆网络(LSTM)是在递归神经网络(RNN)框架下进行的改进算法,解决了传统RNN在处理长序列数据时梯度消失和梯度爆炸问题(Hochreiter et al.,1997)。该算法在处理时间序列数据时具有显著优势,能够有效捕捉长期依赖关系,其记忆单元可自主学习关键历史信息的保留与遗忘,能够更好地适应锚索预应力随服役时间损失的非线性累积特性,对温度、降雨等环境因素的时序波动具有稳健性,从而实现对预应力损失随时间的动态演化过程的精准建模。因此,相较于其他传统机器学习方法,LSTM模型尤其适用于锚索预应力预测问题。
ChengJ, De WaeleW,2024.Weighted average algorithm:A novel meta-heuristic optimization algorithm based on the weighted average position concept[J].Knowledge-Based Systems,305:112564.
[2]
FangBoyang, ZhaoGuoyan, MaJu,et al,2023.Prediction study on loosening ring of surrounding rock around roadways using the optimized ensemble learning algorithms based on Adaboost[J].Gold Science and Technology,31(3):497-506.
[3]
FengZhongju, JiangGuan, ZhaoRuixin,et al,2021.Study on pre-stress long term loss of anchor cable considering coupled multiple factors[J].Rock and Soil Mechanics,42(8):2215-2224.
[4]
GaoX H, TianW P, LiJ C,et al,2021.Research on prediction model of prestress loss of anchor cable in soil-rock dual-structure slope[J].Advances in Materials Science and Engineering,(1):4986245.
[5]
GaoX, JiaJ Q, MeiG X,et al,2022.A new prestress loss calculation model of anchor cable in pile-anchor structure[J].Mathematics,10(8):1260.
JiQiwei, ShiKeyou, HongYingwei,2022.Overall stability analysis of anchorage slope based on prestress monitoring[J].Science Technology and Engineering,22(34):15269-15276.
[8]
LiYingyong,2008.Study on Long Term Stability of Geotechnical Prestressed Anchor System[D].Beijing:Beijing Jiaotong University.
[9]
LiZhenyang, ZhangBaogang, XiongXin,et al,2024.Research on PPV prediction model of open-pit mine based on PSO-XGBoost[J].Gold Science and Technology,32(4):620-630.
[10]
RenQingyang, PengYang, HanKai,et al,2020.Study on anchorage characteristics of prestressed anchor cable under unloading[J].Metal Mine,49(8):61-68.
[11]
SongMinggang, XiangXin, ZhangGuangcheng,et al,2022.Study on mechanical characteristics and prestress loss of slope-anchor structure[J].Chinese Journal of Rock Mechanics and Engineering,41(Supp.1):2791-2800.
[12]
SunShiguo, JiaXinxin, XiaoJian,2022.Research status and development trend analysis of geotechnical pre-stressed anchoring technology[J].Safety in Coal Mines,53(3):213-220.
[13]
WangBenhao, WangYanxia, XiangXueyong,et al,2024.Monitoring and prediction of ground subsidence in mining areas using DS-InSAR and LSTM[J].Bulletin of Surveying and Mapping,(11):44-48.
[14]
WangGuofu, LiQiang, LuLinhai,et al,2018.Research and application of a new coupled model of anchor prestress loss and rock creep[J].Building Structure,48(15):110-114.
[15]
WangJiansong, YuQiangshan, LiuQingyuan,et al,2019.Study on influence of prestressed anchor cable tensioning mode on prestress loss[J].Journal of Highway and Transportation Research and Development,36(9):37-42.
[16]
WangZ Z, DaiG L, LiZ D,et al,2024.Research on H-3K creep coupling calculation model of long-term prestress loss of anchor cables[J].International Journal of Geomechanics,24(2):04023282.
[17]
WuXiaoping, ZhouQuan, HeXuanyi,et al,2023.A prestress loss model of anchor cables for high cutting slopes considering unloading[J].Journal of Central South University(Science and Technology),54(6):2141-2149.
[18]
XieRaoqing, ChenJianhong, XiaoWenfeng,2022.Prediction method of stope stability based on NPCA-GA-BP neural network[J].Gold Science and Technology,30(2):272-281.
[19]
XuYiqing, DengShaoyu, GeQi,2020.Prediction models for short-term and long-term pre-stress loss of anchor cable[J].Rock and Soil Mechanics,41(5):1663-1669.
[20]
YangWendong, LiuChuntian, ZhangXiang,et al,2023.Coupling analysis between prestress loss of anchor cables and creep of rocks[J].Hazard Control in Tunnelling and Underground Engineering,5(2):33-41.
[21]
YuQiangshan, WangJiansong, LiuQingyuan,et al,2019.Long-term prestress loss law and prediction model of anchor cable in strongly weathered granite slope[J].Railway Engineering,59(7):93-96.